Abstract

Enhanced wind datasets of the European satellite Meteosat are now provided every 90 mins together with the quality indicator (QI) derived by the quality control of the Meteorological Product Extraction Facility (MPEF) at the European Organisation for the Exploitation of Meteorological Satellites. All three channel cloud motion winds and clear sky water vapor motion winds have been passively monitored by comparison with the European Centre for Medium-Range Weather Forecasts model background field. The evaluation of the relationship between the MPEF QI and the observation − background differences indicate possible benefits to be gained from the use of the QI within the observation screening of the assimilation system. The MPEF quality indicator is used as a selection criterion within the screening. The applied thresholds are restricted in the Tropics compared to the extratropical regions where the threshold for high-level winds has been relaxed below the automatic quality control at MPEF. The wind data derived from imagery of both Meteosat platforms at 0° and 118°E are used in this study. The overall effect is an increase of active Meteosat winds by a factor of 2. This means a considerably increased impact of Meteosat winds on the tropospheric analyses. The assessment of mean wind increments indicates that the increased temporal sampling together with the use of the quality indicator within the observation screening leads to an improvement of the consistency of the atmospheric motion wind data actively used within the four-dimensional variational assimilation system. The averaged impact on the short- and medium-range forecasts is found to be neutral in the Northern Hemisphere and positive in the Southern Hemisphere. In a selected synoptic case study the use of the new Meteosat wind product indicates a considerable improvement of the medium-range forecasts for the North Atlantic and European areas.

1. Introduction

Research on the extraction of wind fields derived from image sequences taken by geostationary satellites has considerably increased the spatial and temporal coverage of this data type. One major motivation for this development has been the demand of numerical weather forecasting centers for increased coverage of the existing atmospheric motion wind (AMW) datasets and the exploitation of new channels in order to fill data-void areas (Kelly 1993; EUMETSAT 1996). This effort led to improvements on various aspects of the extraction schemes used to derive AMW observations from satellite image loops (Nieman et al. 1997; Ottenbacher et al. 1997; Rattenborg 1998). The resolution of the tracer selection was increased, which provides a larger number of potential tracers. The actual pattern matching task has not been changed significantly. However, the existing scheme can be successfully applied to determine the displacement of cloud tracers as well as cloud-free features within water vapor (WV) imagery. One main change in the operational data stream was the transition to a wind product from the U.S. Geostationary Operational Environmental Satellite (GOES) with increased spatial resolution in March 1998. The higher spatial resolution was followed by the extension to 3-h time sampling compared to the previous 6-h sampling. Since 1997 the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) has been providing an enhanced AMW dataset at 90-min temporal sampling. This includes the traditional cloud motion winds (CMW) derived from cloud tracers observed in the visible (VIS) channel, the infrared (IR) window, and the WV channel. Additionally, the displacement of clear sky features observed in the WV channel are extracted. These are generally referred to as water vapor motion winds. The new dataset is provided simultaneously to the old CMW observations with 6-h time sampling. Since the start of the Indian Ocean Experiment satellite winds from imagery of Meteosat-5 located at 63°E are providing the same coverage over central Asia and the Indian Ocean as the 0° Meteosat mission (Rattenborg and Elliott 1998). This includes the 90-min temporal sampling and cloud drift winds from the VIS channel at full resolution. As a result the total number of AMW vectors received at the European Centre for Medium-Range Weather Forecasts (ECMWF) increased by about one order of magnitude between 1998 and 2000.

One step in the extraction of AMW observations is the quality control prior to the transmission of the data. Traditionally this includes a manual quality check based on a subjective comparison of the wind vector with the actual image triplet used in the retrieval. Following the increase in temporal and spatial resolution the task of a manual control became impossible. It has been recognized earlier by NWP centers that increasing temporal and spatial coverage of observations should be accompanied by the derivation of flags for individual observations in order to enable screening decisions according to certain meteorological situations, the quality of tracers used, or the confidence given to a particular datum during the retrieval process (Kelly 1993). Therefore, the producers are working on the development of quality estimates for individual wind observations simultaneously to the attempts to increase the spatial and temporal coverage (Hayden and Purser 1995; Holmlund 1998; Holmlund and Velden 1998). The new data stream from EUMETSAT contains all observations that pass a very weak quality threshold of QI > 0.3 together with the final quality indicator (QI) assigned during the quality control at the Meteorological Product Extraction Facility (MPEF).

The cloud-tracked winds generally provide an inhomogeneous dataset. The revised extraction scheme not only increases the density in areas already covered but is also capable of providing new information in previous data voids (Rohn et al. 1998). In areas with a large number of suitable tracers a dense set of wind observations is derived. This is a common situation for high cirrus clouds along the jet streams in the midlatitudes. Since a large number of observations are derived from similar cloud systems the errors of the actual tracking step as well as the height assignment can be expected to be correlated. Together with the increased number of observations the numerical analysis becomes more sensitive to error sources. One importance for cloud-tracked winds remains the underestimation of high wind speeds although it has been considerably improved by the satellite winds producers during recent years (Kållberg 1997). The increased number of AMW data therefore requires more care in the assimilation of this observation type. Following the increased density of satellite winds from both GOES platforms a thinning of cloud-tracked winds was introduced in the observation screening at ECMWF in 1998. The thinning of satellite winds ensures a minimum horizontal and vertical distance between observations. An impact study performed at ECMWF in order to assess the relative impact of the conventional and the increased density of GOES winds did not show an immediate benefit to be gained from the increased resolution. A thorough study at that time was complicated by a restricted availability of concurrent datasets with different resolution. However, it indicated that there is no obvious improvement of analyses and forecasts by only increasing the density of the observations. The introduction of the thinning was a precautionary measure in order to reduce the effect of correlated observation errors in areas of dense coverage by adapting the density of active observations to the resolution of the assimilation.

In this paper we present the introduction of the new AMW data with 90-min time sampling from Meteosat into the four-dimensional variational (4DVAR) assimilation system at ECMWF (Rabier et al. 2000; Klinker et al. 2000). We restricted the study to cloud motion winds since the clear sky water vapor motion vectors represent very different information that is currently under investigation (Kelly et al. 1998). In particular the water vapor motion vectors should be assessed in the context of the current work on the assimilation of geostationary radiances by Munro et al. (1998). The QI derived during the quality control of AMW observations from Meteosat is used as a selection criterion within the observation screening. In the first section we show an empirical assessment of the QI value by monitoring the relation between the QI value and the departures of the AMQ data from the NWP forecast fields. From these findings we propose possibilities to use the QI values during the screening of observations in a numerical data assimilation system. The approach has been applied to satellite winds derived during the North Pacific Experiment (Langland et al. 1999; Holmlund et al. 2001). We discuss the results with the use of the QI values in assimilating the Meteosat winds with the ECMWF 4DVAR system. The details of the applied screening decisions are described together with the results from assimilation and forecast experiments. This system was introduced into the operational ECMWF model assimilation in July 1999.

2. Monitoring of the MPEF quality indicator

The MPEF QI value is traditionally used as a criterion by the automatic quality control (AQC) to decide about the transmission of a particular datum to users. The current AQC setup includes only observations that exceed a threshold of QI = 0.8 in the 6-hourly CMW dataset except for CMW observations from the visible imagery at high resolution where a threshold of QI = 0.65 is applied. A concise description of the MPEF quality control scheme and the derivation of the QI value can be found in Holmlund (1998). The quality of new MPEF winds have been routinely compared against the ECMWF background wind field for several years. The background field is a 6-h forecast of wind and pressure with 1.5° horizontal resolution at 50 vertical model levels. In particular the relation between the background departures (i.e., difference between observed and 6-h forecast) and the assigned QI have been monitored in order to assess the information content of the QI values toward a possible use within an assimilation system. Following the monitoring practice at MPEF the rms background departures versus assigned quality estimate are studied separately for different channels (IR, VIS low and high resolution, WV cloud, and clear sky), in different geographical regions, and tropospheric layers. Due to the problem of negative speed bias in high wind speed situations the mean wind speed and bias are included in this monitoring. The relation between QI value and background departures for high level (p < 400 hPa) IR winds is shown in Fig. 1. The statistics have been computed for a one-week period from 1 to 7 February 1999. The departures are collected in QI bins of 0.01. All cloud-tracked winds as well as the motion vectors from clear sky features within WV images have been assessed separately for the northern (latitude > 20°) and southern (latitude < −20°) extratropical regions, and the Tropics (−20° < latitude < 20°). Additionally three tropospheric layers are distinguished (p > 700 hPa, 700 hPa > p > 400 hPa, 400 hPa > p). As an example of low-level winds in the Tropics the monitoring of the high-resolution VIS winds is shown in Fig. 2. The operational monitoring of the wind observations by comparison against collocated radiosonde wind measurements has shown a monotonic decrease of rms wind vector departures normalized by the wind speed with increasing QI value (Schmetz et al. 1993). Generally, the background statistic shows decreasing rms departures and bias with an increasing quality estimate (QI). Furthermore, the mean wind speed in each quality interval is also increasing with the quality indicator. These findings are supported by the results for the monitoring of the quality indicator presented in earlier studies (Rohn et al. 1998; Holmlund et al. 2001). This is important with respect to the underestimation of high wind speeds by cloud-tracked winds. Within the ECMWF assimilation system the so-called asymmetric check is applied as part of the first guess check of atmospheric motion vectors in order to prevent a slowing down of jet streams (Järvinen and Undén 1997). This background check is applied to observations that are slower than the background wind speed by 4 m s−1. The details of the parameter choice are summarized by Tomassini et al. (1999). The observed relation between quality indicator and background departures indicates the potential of the QI value as an additional parameter within observation screening of a NWP assimilation system.

Fig. 1.

Monitoring of high-level (p < 400 hPa) IR winds from Meteosat-7 (1–7 Feb 1999). The rms departures are marked by the solid line, the speed bias in dotted style, and the background wind speed as dashed line. The histogram shows the number of observations in each QI bin

Fig. 1.

Monitoring of high-level (p < 400 hPa) IR winds from Meteosat-7 (1–7 Feb 1999). The rms departures are marked by the solid line, the speed bias in dotted style, and the background wind speed as dashed line. The histogram shows the number of observations in each QI bin

Fig. 2.

Monitoring of tropical low-level (p > 700 hPa) high-resolution VIS winds from Meteosat-7 (1–7 Feb 1999). The rms departures are marked by the solid line, the speed bias in dotted style, and the background wind speed as dashed line. The histogram shows the number of observations in each QI bin

Fig. 2.

Monitoring of tropical low-level (p > 700 hPa) high-resolution VIS winds from Meteosat-7 (1–7 Feb 1999). The rms departures are marked by the solid line, the speed bias in dotted style, and the background wind speed as dashed line. The histogram shows the number of observations in each QI bin

3. Potential of enhanced satellite winds

The new dataset generally provides more information due to increased spatial and temporal resolution as well as including the winds that have been assigned a quality mark below the automatic quality control threshold. Using a case study we illustrate the potential of the extended observation set. Figure 3 shows mid- and upper-troposheric winds from Meteosat-7 at 1200 UTC 19 March 1999. The vectors shown are observations that passed the AQC (black) and the enhanced satellite winds (red). Multiple observations at some locations represent data from tracking of cloud features within different imaging channels. This is in contrast to the traditional wind product, which includes only one wind per processing segment, which is identified by the highest QI value. Therefore more than one observation can be provided within the same segment by tracking of clouds at different levels. Motion vectors derived in clear sky areas of WV imagery are not included. Figure 3 illustrates only the potential of cloud-tracked winds. The “fleet” of observations over North Africa where no clouds can be identified on the satellite image (not shown) are derived from the WV channel and indicate tracking of either thin cirrus at upper levels or uncondensed moisture, which was incorrectly classified by the cloud detection scheme. The comparison with the geopotential analysis at 400 hPa subjectively gives confidence in the additional observations. There is a good indication that the quality control rejects many observations that appear to supplement information on the vector field in a consistent way. This is especially clear in the vicinity of the low pressure system over the Mediterranean. A few obvious outliers are also apparent, which is to be expected due to the relaxed quality control. One reason for low confidence of the vectors close to the center of the low is higher rejections due to the spatial consistency test in regions of strong flow curvature. This is explained by the spatial consistency check between neighboring wind vectors, which has double weight within the final result of the quality control (Holmlund 1998). Another reason can be reduced confidence from the forecast consistency check. The best background provided to EUMETSAT is a 12-h forecast field. Especially in the case of mesoscale features small differences in the location of the low center or the region of strong gradients can cause considerable differences in the results of the forecast consistency check. This is an unlikely situation within a data-rich area like Europe where analysis and subsequent 12-h forecast are of general high quality. However, it becomes important over data-void regions and consequently of particular importance for the current Meteosat-5 data covering the Indian Ocean. In the context of data assimilation for numerical weather prediction it is therefore desirable to remove as much dependency of the observations from the NWP background field as possible.

Fig. 3.

Atmospheric motion winds from Meteosat-7 between 600 and 200 hPa at 1200 UTC on 19 Mar 1999. The old observations (6-h winds passing the AQC) are marked by black flags while the new data (only nearest time slot at 1230 UTC) are plotted in red. All observations are overlayed on the analysis of the geopotential height at 400 hPa at 1200 UTC

Fig. 3.

Atmospheric motion winds from Meteosat-7 between 600 and 200 hPa at 1200 UTC on 19 Mar 1999. The old observations (6-h winds passing the AQC) are marked by black flags while the new data (only nearest time slot at 1230 UTC) are plotted in red. All observations are overlayed on the analysis of the geopotential height at 400 hPa at 1200 UTC

4. Use of QI for screening of atmospheric motion vectors

a. Blacklisting

The blacklist provides a flexible means of excluding observations on the basis of continuous data monitoring. In contrast to background checks that are implemented within the assimilation system the blacklist allows a quick response to observed changes in the data quality and allows thresholds based on the MPEF quality indicator to be used. Based on the monitoring described a set of QI thresholds is derived for blacklist decisions for the three-channel cloud-tracked winds. The parameter choice is empirical as is the AQC threshold of 0.8 used for the dissemination of winds at MPEF. Quality is generally measured by comparison to other sources of wind data, the most independent one being radiosonde measurements. Their availability is generally sparse and differs geographically. For example it is hard to verify satellite winds over the Atlantic or Indian Oceans only by sonde observations except during a special field campaign. However, these are the most important target areas for the application of satellite observations. An alternative quality monitoring is performed by the comparison to the analysis and background field of a numerical weather prediction system. This verification however can never be purely objective since the background is influenced by the observations. Additionally, the processing of the observations themselves involves the use of analysis or forecast fields (e.g., height assignment). In tuning any quality thresholds one attempts to balance two opposing objectives:

  • minimization of the error of the observation as measured by rms departures and bias on a statistical average—for example, the routine data monitoring at ECMWF is done on a three 3-month basis, and

  • since the rms departures and bias arise both from observation and model errors a certain level of rms departures has to be retained in order to supply new information and to correct the numerical model background field.

In this study we chose relaxed thresholds that increase the coverage allowing for increased rms departures by 20% compared to the statistics observed for the conventional Meteosat winds. This initial arbitrary choice was then used for impact studies. For observations within the tropical region the thresholds were tightened again following the experience from an impact study. Unlike the MPEF quality control scheme, the cutoff parameters are chosen separately for channel, tropospheric layer, and geographical region. The selection that was used in the assimilation experiments described later (ECQC-90) is summarized in Table 1. Generally, it led to a restriction in the use of low-level IR cloud-tracked winds, and at midlevel IR winds were activated about QI > 0.90. The latter winds were not used within the operational system based on the background departures observed in the past (Rattenborg 1998). The threshold for high-level winds has been relaxed down to QI > 0.60 in the extratropical region while the usage for tropical winds is more restricted compared to the MPEF AQC. For VIS winds the AQC decision has been used. The VIS winds at high resolution are provided three hourly together with the low-resolution VIS winds every 90 min. The time slots of low-resolution VIS data that coincide with the high-resolution VIS winds are blacklisted to avoid redundancy. These decisions are applied in conjunction with the transition to 90-min time sampling. The overall effect can be expected to be a large increase in the number of satellite winds being presented to the analysis. The main reasons in order of decreasing importance are 90-min time sampling, a relaxed QI threshold at high level in the extratropical region, multiple observations per processing segment from different channels at identical time slots, and the introduction of IR winds at medium levels.

Table 1.

Data selection according to quality estimate QI

Data selection according to quality estimate QI
Data selection according to quality estimate QI

b. Selection in thinning

Besides the static QI thresholds the thinning step for atmospheric motion vectors has been extended using the MPEF QI value. The modification is such that conventional cloud-tracked winds without quality flags are separated from those with individual quality estimates. The selection follows the thinning scheme applied to various observations types with high spatial coverage (Järvinen and Undén 1997). Atmospheric motion vectors are collected in boxes of dimension as generally used for all CMW observations (Table 2). This ensures a minimum horizontal distance of 1.25° between AMW data within the same pressure layer. The pressure layer is defined by the model pressure level nearest to the observed AMW pressure height. The quality estimate QI is included as the selection criterion within each thinning box. In the presence of several observations within one thinning box the observation with the highest quality estimate is retained as active in the assimilation. In the case of identical QI values the smallest difference between observational and analysis time is preferred. Since the 4DVAR screening is performed in 1-h time steps this situation can only occur with coinciding AMW observations from different Meteosat platforms (i.e., at 0° and 63°E).

Table 2.

Specifications of satellite observation thinning box. Only one observation is retained as active depending on QI value and the distance between the observational and the analysis time

Specifications of satellite observation thinning box. Only one observation is retained as active depending on QI value and the distance between the observational and the analysis time
Specifications of satellite observation thinning box. Only one observation is retained as active depending on QI value and the distance between the observational and the analysis time

5. Assimilation and forecast experiments

The experiment hereafter referred to as ECQC-90 introduces the full 90-min sampling using the QI thresholds based on the monitoring described in section 2 (Table 1). Two experiments were performed using the ECMWF four-dimensional assimilation system (21 Oct–10 Nov 1998 and 6–31 May 1999). Clear sky WV winds were not used in either the operational suite or the experiments. Both experiments are compared to the four-dimensional assimilation system that was operational during the relevant time period. The main difference between the experiments is the extension of the vertical resolution into the stratosphere (50 levels compared to 31 levels). This can be expected to be of minor importance in this context due to the similar vertical resolution between the surface and 50 hPa. The operational system (Control) uses only cloud-tracked winds at the synoptic times (0000, 0600, 1200, and 1800 UTC) that passed the automatic quality control at MPEF. See Tomassini et al. (1999) for more details of the AMW usage. The main differences in the data usage are therefore the increased time sampling, the revised usage of the winds, and the introduction of the QI value as selection criterion as outlined above (and Table 3).

Table 3.

Experimental setup for tuning of QI thresholds and transition to 90-min winds. The values during both periods (21 Oct–10 Nov 1998 and 6–31 May 1999) are identical

Experimental setup for tuning of QI thresholds and transition to 90-min winds. The values during both periods (21 Oct–10 Nov 1998 and 6–31 May 1999) are identical
Experimental setup for tuning of QI thresholds and transition to 90-min winds. The values during both periods (21 Oct–10 Nov 1998 and 6–31 May 1999) are identical

a. Analysis/Impact

The background and analysis departures of the CMW observations used are discussed first for the experiment ECQC-90 during the November 1998 period. The statistics include 10 assimilation cycles at 1200 UTC during the period 1–10 November 1998. The results are confined to the area covered by both Meteosat platforms (55°S < latitude < 55°N, 55°W < longitude < 118°E). The number of active CMW winds in the analysis has been increased by a factor of 2 compared to the operational analysis (Table 4). Differences in the number of active observations during the second experiment in May 1999 are also summarized in Table 4 for all assimilation cycles during the 26-day period.

Table 4.

Number of active CMW observations for both experiments (55°S < latitude < 55°N, 55°W–118°E)

Number of active CMW observations for both experiments (55°S < latitude < 55°N, 55°W–118°E)
Number of active CMW observations for both experiments (55°S < latitude < 55°N, 55°W–118°E)

The vertical distribution of rms departures and bias of CMW observations from the background and analysis fields are plotted in Fig. 4 for the November 1998 experiment. The rms background and analysis departures of active CMW observations have been considerably decreased at all levels. A pronounced change of the zonal wind component bias occurs at midtropospheric levels where no CMW winds from Meteosat were used in the operational analysis. The midlevel observations used in the operational analysis are IR winds from GOES-8 that partly cover the Meteosat area between 55° and 15°W over the Atlantic. Since CMW observations are generally blacklisted over land below 500 hPa the differences must represent a different regime over the Indian Ocean. The fit of other observations to background and analysis within the area of influence of CMW data from Meteosat has hardly changed. Small differences in the fit of pilot sonde wind observations show decreased rms departures and bias especially at upper levels (not shown). The fit of active CMW observations during the May 1999 experiment also shows a closer fit by decreased rms departures. The bias of the zonal component is decreased at midtropospheric levels and increased at high levels. Note that in the extratropical region many more winds are presented to the analysis at high levels (QI > 0.6 for p < 400 hPa), which on average show larger departures from the background than above the AQC threshold (QI > 0.8). See for instance the monitoring of high-level winds in Fig. 1. The bias and rms departures of the CMW meridional component are both decreased. Again the fit of active radiosonde observations remains mainly unchanged. A few additional sonde data are activated at low- and upper-tropospheric levels and more rejections occur at medium levels and in the lower stratosphere. These changes of quality control decisions of the analysis system are very small (≈0.01%) and are due to changes in the background field. Minute differences in the background departures indicate an improved rms fit. The different signal in the fit of satellite winds and other wind observations indicates that the observations are influencing different areas. This agrees with the geographical distribution of the mean rms increments discussed below.

Fig. 4.

Departures of active AMW observations from background and analysis. The Control analysis is marked by dashed–dotted line, and the Control background by dashed lines. Analysis and background departures of the ECQC-90 experiment are marked in solid and dotted, respectively. The differences in the number of observations refer to the experiment using the new ECQC-90 setup. The statistics are collected within an area covered by both Meteosat platforms (−55° < latitude < 55°, −55°W to 118°E) during 1–10 Nov 1998 at 1200 UTC

Fig. 4.

Departures of active AMW observations from background and analysis. The Control analysis is marked by dashed–dotted line, and the Control background by dashed lines. Analysis and background departures of the ECQC-90 experiment are marked in solid and dotted, respectively. The differences in the number of observations refer to the experiment using the new ECQC-90 setup. The statistics are collected within an area covered by both Meteosat platforms (−55° < latitude < 55°, −55°W to 118°E) during 1–10 Nov 1998 at 1200 UTC

Changes in the increments of the geopotential height are further used in order to assess the quality of the modifications introduced into the analysis. The increment is defined as the difference between the analysis and the 6-h forecast for the same time. The differences of rms increments are averaged over the entire experimentation period. In the following we discuss the difference field of rms increments between the experiment assimilating Meteosat winds with the ECQC-90 setup and the Control using the operational Meteosat winds with 6-h time sampling provided by the MPEF quality control. The resulting fields for the geopotential at 250 hPa are presented in Fig. 5. The results for the autumn and the spring experiment are shown in the top and bottom panel, respectively. First of all the area of impact for Meteosat winds from both platforms is clearly visible. The observations are generally used only within a great circle of 55°. The rms increments are reduced within extended areas over the oceans especially in the Southern Hemisphere. The spring experiment (bottom panel) reveals increased rms increments right on and beyond the data boundary of 55°S latitude. In the Northern Hemispher satellite winds are traditionally not used over land with the exception of North Africa (Tomassini et al. 1999). This restriction has been relaxed with the introduction of observations from Meteosat-5 over the Indian Ocean beyond 30°E where AMW observations are active within the upper troposphere (p < 500 hPa). Consequently, the increment differences reveal no changes over Europe. In the area directly influenced by Meteosat winds the rms increments are reduced with a strong impact being centered over Kazakhstan. It is not obvious that an increase in active observations by a factor of 2 leads to reduced increments. This indicates that the changes in the observational system are mutually consistent and also with the model background, and other observations within the same area (E. Andersson, ECMWF, 1999, personal communication). Increased increments can indicate disagreement between satellite winds themselves, the background, or other observations. Television Infrared Observation Satellite (TIROS) Operational Vertical Sounder TOVS radiances from tropospheric sounding channels are not used over land in either of the experiments. The remaining alternative measurements over Asia are therefore aircraft and sonde data. The strongest reduction of mean increments coincides with a data-sparse area over Kazakhstan. The differences in the rms increments at 850 hPa (not shown) reveal reductions only in the Southern Hemisphere coinciding with the impact on the 250-hPa field. The absence of significant impact in the Northern Hemisphere is consistent with the fact that AMW data are generally not used over land below a pressure height of 500 hPa. The large areas of reduced increments indicate that the analysis has been improved toward a system that is mutually consistent with respect to background and observational information. It should be noted that the relaxed QI threshold (QI > 0.6) for upper-level observations (p < 400 hPa) in the extratropical areas presents many observations with increased background departures to the analysis. The four-dimensional assimilation appears to make good use of the 90-min AMW data. An interaction between the first guess check, the variational quality control, and the screening decisions dependent on the QI value also contributes to this positive result.

Fig. 5.

Differences in mean rms increment (×10 gpm) of the geopotential height at 250 hPa between the experiment assimilating 90-min winds and QI dependent screening (ECQC-90) and the operational analysis as control: (top) 21 Oct–10 Nov 1998, (bottom) 6–31 May 1999. Negative values (marked in green) indicate reduced rms increments. Positive values are marked in yellow. The contour lines show the mean analyzed geopotential field over the experimentation period

Fig. 5.

Differences in mean rms increment (×10 gpm) of the geopotential height at 250 hPa between the experiment assimilating 90-min winds and QI dependent screening (ECQC-90) and the operational analysis as control: (top) 21 Oct–10 Nov 1998, (bottom) 6–31 May 1999. Negative values (marked in green) indicate reduced rms increments. Positive values are marked in yellow. The contour lines show the mean analyzed geopotential field over the experimentation period

b. Forecast impact

The assessment of the development of short-term forecast errors between 24 and 72 h shows the main signal in the Southern Hemisphere. Figure 6 contains the rms errors difference between the fall experiment (ECQC-90, 21 Oct–7 Nov 1998) and the operational forecast of the 250-hPa geopotential height. Light gray shading marks negative values and improved forecast. Dark gray shading indicates increased rms forecast errors. The 24-h forecast (top-left panel) reveals extended areas of reduced forecast errors. During the next two steps (top right, 48 h; bottom, 72 h) the main contribution appears to concentrate in the South Indian Ocean. The spring experiment (6–31 May 1999) shows a similar pattern with some areas of increased rms errors in the South Indian Ocean and an area of reduced errors farther west between 20°W and 20°E. During the spring experiment the verification of the medium-range forecast in the Northern Hemisphere is strongly influenced by a single synoptic event. The impact is mainly neutral up to 16 May. Simultaneously with a drop in the skill of the operational suite during the period 17–19 May 1999 the experiment also verifies worse, resulting on average in a negative impact in the Northern Hemisphere. A further reduction of anomaly correlation from 40% to 30% does not provide sufficient confidence in this signal. This is addressed by an additional cross-check together with the case study below. In order to address the sampling problem the forecast impact has been assessed for both experiments (fall and spring) together. The evaluation of the forecast skill for geopotential was based on verification against the Control analysis (operational system), which includes all 47 forecast cases (21 Oct–11 Nov 1998 and 6–31 May 1999). The overall impact is neutral in the Northern Hemisphere and positive in the Southern Hemisphere. The wind forecasts at 850 and 200 hPa for both hemispheres and for the Tropics have been verified against their own analyses (33 Cases). The errors are reduced in the Southern Hemisphere at both levels. An initial test with the 90-min winds using the same QI thresholds for tropical and extratropical winds showed increased rms errors for the wind forecast in the Tropics. The introduction of frequent observations every 90 min in the Tropics therefore appeared to require more care than in the extratropical regions. This led to the restricted QI thresholds given in Table 1. A possible explanation is the larger impact of AMW data on the tropical analysis. A good example is the strengthening of the cross-equatorial flow during an active monsoon observed after the positioning of Meteosat-5 over the Indian Ocean (Lalaurette et al. 1998). The wind forecast errors in the Tropics show a small decrease at 850 hPa at day 4 and a slight increase at 200 hPa beyond day 4. The significance of the differences in forecast skill between ECQC-90 and the Control run was studied using the Student's t-test. Different significance levels were applied to test the basic hypothesis of the t test: the averaged errors (or the averaged anomaly correlation) of both samples are equal. The test was generally applied to the forecast of geopotential height at 1000, 500, and 200 hPa and vector wind at 850 and 200 hPa. The significance of differences in forecast skill was studied separately within the Northern Hemisphere, Tropics, and Southern Hemisphere. The results are summarized in Tables 5 and 6 for geopotential and vector wind, respectively. Only results below or equal an error level of 5% were included. The statements “better” or “worse” mean that the basic hypothesis can be ruled out with the probability of being wrong as given by the significance level.

Fig. 6.

Difference in mean rms forecast error for the geopotential height at 250 hPa verified against own analysis (21 Oct–7 Nov 1998): (top left) t + 24 h, (top right) t + 48 h, and (bottom) t + 72 h. Light-gray areas mark decreased forecast errors. The shading is restricted to differences exceeding (±) 0.2, 0.5, and 1.0 m, respectively

Fig. 6.

Difference in mean rms forecast error for the geopotential height at 250 hPa verified against own analysis (21 Oct–7 Nov 1998): (top left) t + 24 h, (top right) t + 48 h, and (bottom) t + 72 h. Light-gray areas mark decreased forecast errors. The shading is restricted to differences exceeding (±) 0.2, 0.5, and 1.0 m, respectively

Table 5.

Significance of differences in forecast quality between ECQC-90 and the Control run for the forecast of geopotential height (21 Oct–11 Nov 1998 and 6–31 May 1999). The significance of the differences in anomaly correlation and rms error is studied using the Student's t-test. The statement “Better” or “Worse” means that the basic hypothesis of on average equal skill can be ruled out with the probability of being wrong to the applied significance level

Significance of differences in forecast quality between ECQC-90 and the Control run for the forecast of geopotential height (21 Oct–11 Nov 1998 and 6–31 May 1999). The significance of the differences in anomaly correlation and rms error is studied using the Student's t-test. The statement “Better” or “Worse” means that the basic hypothesis of on average equal skill can be ruled out with the probability of being wrong to the applied significance level
Significance of differences in forecast quality between ECQC-90 and the Control run for the forecast of geopotential height (21 Oct–11 Nov 1998 and 6–31 May 1999). The significance of the differences in anomaly correlation and rms error is studied using the Student's t-test. The statement “Better” or “Worse” means that the basic hypothesis of on average equal skill can be ruled out with the probability of being wrong to the applied significance level

c. Cross-check and case study

The experimentation during the spring period (6–31 May 1999) was the final test before the introduction into a new operational assimilation and forecast system (Esuite). This system serves to assess finally the combined impact of different modifications compared with the operational system at the time (Osuite). The change in the operational use of Meteosat winds (ECQC-90) was included together with other changes in the observational system. An additional cross-check was performed in order to address the importance of the ECQC-90 modifications during the bad forecast period mentioned above. This test consisted of simply reactivating the former usage of Meteosat winds within the complete new operational package (Esuite “minus” ECQC-90). The investigation of the synoptic situation revealed that the operational system forecast a pronounced ridge instead of the analyzed zonal flow over the North Atlantic as well as misplacing a low pressure system east of Iceland. Figure 7 shows the analysis field (top left) together with the operational forecast (top right). The new system (Esuite) shows a considerably improved forecast in this particular synoptic situation (bottom left). Finally, disabling the ECQC-90 contribution within the Esuite (Esuite minus ECQC-90) almost entirely reproduces the previous failure of the operational forecast (bottom right). Therefore, a direct connection between the bad forecast and AMW data can be ruled out. Moreover, the modified use of Meteosat winds is found to provide an important positive impact on the forecast in this particular synoptic situation. A possible explanation is the interaction of the AMW data with other new components of the observational system. The Esuite also introduced a revised background check for radiance observations of the Microwave Sounding Unit and Advanced Microwave Sounding Unit-A instruments. The rejection limits were relaxed leading to an increased number of these observations. A positive interaction between satellite winds and radiance data has been reported earlier in a study on the influence of different observation types on the ECMWF forecast system (Kelly 1997). The presence of a particular observational component might avoid useful information being rejected by the model. This issue has been studied in the context of assimilating asynoptic observations as temperature or geopotential information derived from spaceborne instruments. The problem is illustrated in the context of continuous data assimilation by Daley (1991). He describes how the assimilation of geopotential can be supported by providing wind increments that correspond to the geopotential observation through the geostrophic relation.

Fig. 7.

The 5-day forecasts of the geopotential height at 500 hPa of different systems verifying on 22 May 1999: (top left) operational analysis, (top right) operational forecast (Osuite), (bottom left) complete new suite including the ECQC-90 contribution (Esuite), and (bottom right) new suite without the ECQC-90 contribution (Esuite minus ECQC-90)

Fig. 7.

The 5-day forecasts of the geopotential height at 500 hPa of different systems verifying on 22 May 1999: (top left) operational analysis, (top right) operational forecast (Osuite), (bottom left) complete new suite including the ECQC-90 contribution (Esuite), and (bottom right) new suite without the ECQC-90 contribution (Esuite minus ECQC-90)

6. Summary

The new winds from both Meteosat platforms at 0° and 63°E have been monitored especially with respect to the information content of the MPEF QI quality estimate toward its possible use for data selection within an NWP system. The QI value appears to be capable of specifying the quality regarding rms wind vector error, speed bias, and the observed wind speed. This provides good evidence to include the QI value as an additional criterion in the observation screening and so a set of refined QI thresholds has been derived from the monitoring. The QI value of active observations is further used as a selection criterion within the thinning step for AMW data. This scheme is applied to Meteosat winds with a 90-min time sampling. The overall effect in the analysis is an increase in the number of active CMW data by a factor of 2. The two main reasons are the increased time sampling and a relaxed QI threshold at high level in the extratropical regions. The fit of CMW data to background and analysis has been improved. The fit of other observations remains essentially unchanged. The rms increments for geopotential at 850 and 250 hPa are generally reduced in extended areas over the oceans, North Africa, and Asia. This indicates an improved analysis in terms of the consistency between the active Meteosat wind data, the background, and other available observations. The four-dimensional variational assimilation system together with screening decisions depending on the MPEF quality indicator makes good use of the increased time sampling of AMW data. It was found that the increase of temporal sampling from 6 h to 90 min requires higher QI thresholds in the Tropics compared to the extratropical regions where the QI threshold at upper levels could be relaxed below the MPEF AQC. In a selected synoptic situation the modification in the use of AMW data from both Meteosat platforms appears to provide important benefits for the medium-range forecast over the North Atlantic and Europe. The verification of tropical wind forecasts shows decreased errors in the short term and a small increase beyond day 4. The overall impact on the forecast appears to be neutral in the Northern Hemisphere and positive in the Southern Hemisphere for both short- and medium-range forecasts.

The increase in the number of available AMW observations will continue with the new Meteosat Second-Generation instrument providing images every 15 min. Two WV channels are sensitive to different tropospheric layers. The ozone channel could provide low-stratospheric information (EUMETSAT 1998). It is planned to apply the same approach to AMW datasets from GOES and other geostationary platforms in the future.

Table 6.

Significance of differences in rms forecast error between ECQC-90 and the Control run for the forecast of wind. Both ECQC-90 and the Control run are verified against their own analysis. The results are based on the evaluation of 33 forecast cases (21 Oct–3 Nov 1998 and 6–24 May 1999). The significance of the differences in rms error are studied using the Student's t test. The statement “Better” or “Worse” means that the basic hypothesis of on average equal skill can be ruled out with the probability of being wrong to the applied significance level

Significance of differences in rms forecast error between ECQC-90 and the Control run for the forecast of wind. Both ECQC-90 and the Control run are verified against their own analysis. The results are based on the evaluation of 33 forecast cases (21 Oct–3 Nov 1998 and 6–24 May 1999). The significance of the differences in rms error are studied using the Student's t test. The statement “Better” or “Worse” means that the basic hypothesis of on average equal skill can be ruled out with the probability of being wrong to the applied significance level
Significance of differences in rms forecast error between ECQC-90 and the Control run for the forecast of wind. Both ECQC-90 and the Control run are verified against their own analysis. The results are based on the evaluation of 33 forecast cases (21 Oct–3 Nov 1998 and 6–24 May 1999). The significance of the differences in rms error are studied using the Student's t test. The statement “Better” or “Worse” means that the basic hypothesis of on average equal skill can be ruled out with the probability of being wrong to the applied significance level

Acknowledgments

This work was only possible through support from others in numerous technical and scientific issues. Thanks to all staff and consultants in both the Operations and Research Departments at ECMWF. We would like to thank also Simon Elliott, Ken Holmlund, and Mikael Rattenborg at EUMETSAT for their strong support on the wind production. The research was funded by the EUMETSAT Fellowship Programme.

REFERENCES

REFERENCES
Daley
,
R.
,
1991
:
Atmospheric Data Analysis.
1 ed. Cambridge University Press, 457 pp
.
EUMETSAT
,
1996
:
Summary of the Third International Winds Workshop.
Proc. Third Int. Winds Workshop, Ascona, Switzerland, EUMETSAT Publ. EUM-P-18, xi–xvi
.
EUMETSAT
,
1998
:
SEVIRI science plan.
EUMETSAT Doc. EUM/MSG/PLN/193, 68 pp. [Available online at www.eumetsat.de.]
.
Hayden
,
C. M.
, and
R. J.
Purser
,
1995
:
Recursive filter objective analysis of meteorological fields: Applications to NESDIS operational processing.
J. Appl. Meteor
,
34
,
3
15
.
Holmlund
,
K.
,
1998
:
The utilization of statistical properties of satellite-derived atmospheric motion vectors to derive quality indicators.
Wea. Forecasting
,
13
,
1093
1104
.
Holmlund
,
K.
, and
C.
Velden
,
1998
:
Objective determination of the reliability of satellite-derived atmospheric motion vectors.
Proc. Fourth Int. Winds Workshop, Saanenmoeser, Switzerland, EUMETSAT Publ. EUM-P-24, 215–224
.
Holmlund
,
K.
,
C.
Velden
, and
M.
Rohn
,
2001
:
Enhanced automated quality control applied to high-density GOES winds.
Mon. Wea. Rev
,
129
,
517
529
.
Järvinen
,
H.
, and
P.
Undén
,
1997
:
Observation screening and back ground quality control in the ECMWF 3D-Var data assimilation system.
ECMWF Tech. Memo. 236, 33 pp
.
Kållberg
,
P.
,
1997
:
Aspects of the re-analyzed climate.
ECMWF Re-Analysis Project Report Series, No. 2, 89 pp
.
Kelly
,
G.
,
1993
:
Numerical experiments using cloud motion winds at ECMWF.
Proc. Developments in the Use of Satellite Data in Numerical Weather Prediction, Reading, United Kingdom, ECMWF,
.
331
348
.
Kelly
,
G.
,
1997
:
Influence of observations on the operational ECMWF system.
World Meteorological Organization Bulletin, Vol. 46, No. 4, 336–342
.
Kelly
,
G.
,
R.
Munro
,
M.
Rohn
, and
K.
Holmlund
,
1998
:
Impact of atmospheric motion vectors (AMV) on the ECMWF system and the development of a water vapour AMV observation operator.
Proc. Fourth Int. Winds Workshop, Saanenmoeser, Switzerland, EUMETSAT Publ. EUM-P-24, 125–137
.
Klinker
,
E.
,
F.
Rabier
,
G.
Kelly
, and
J-F.
Mahfouf
,
2000
:
The ECMWF operational implementation of four-dimensional variational assimilation. Part III: Experimental results and diagnostics with operational configuration.
Quart. J. Roy. Meteor. Soc
,
126
,
1191
1216
.
Lalaurette
,
F.
,
A.
Garcia-Mendez
, and
M.
Rohn
,
1998
:
Monitoring SATOB products at ECMWF: Availability, quality control, and impact.
Proc. Fourth Int. Winds Workshop, Saanenmoeser, Switzerland, EUMETSAT Publ. EUM-P-24, 239–246
.
Langland
,
R.
,
and Coauthors,
.
1999
:
The North Pacific Experiments (NORPEX-98): Targeted observations for improved North American weather forecasts.
Bull. Amer. Meteor. Soc
,
80
,
1363
1384
.
Munro
,
R.
,
G.
Kelly
,
M.
Rohn
, and
R. W.
Saunders
,
1998
:
Assimilation of Meteosat radiance data within the 4DVAR system at ECMWF.
Proc. Fourth Int. Winds Workshop, Saanenmoeser Switzerland, EUMETSAT Publ. EUM-P-24,
.
299
306
.
Nieman
,
S. J.
,
W. P.
Menzel
,
C. M.
Hayden
,
D.
Gray
,
S. T.
Wanzong
,
C. S.
Velden
, and
J.
Daniels
,
1997
:
Fully automated cloud-drift winds in NESDIS operations.
Bull. Amer. Meteor. Soc
,
78
,
1121
1133
.
Ottenbacher
,
A.
,
M.
Tomassini
,
K.
Holmlund
, and
J.
Schmetz
,
1997
:
Low-level cloud motion winds from Meteosat high-resolution visible imagery.
Wea. Forecasting
,
12
,
175
184
.
Rabier
,
F.
,
H.
Järvinen
,
E.
Klinker
,
J-F.
Mahfouf
, and
A.
Simmons
,
2000
:
The ECMWF operational implementation of four-dimensional variational assimilation. I: Experimental results with simplified physics.
Quart. J. Roy. Meteor. Soc
,
126
,
1143
1170
.
Rattenborg
,
M.
,
1998
:
Status and development of operational Meteosat wind products.
Proc. Fourth Int. Winds Workshop, Saanenmoeser, Switzerland, EUMETSAT Publ. EUM-P-24, 49–59
.
Rattenborg
,
M.
, and
S.
Elliott
,
1998
:
Meteosat-5 winds for the INDOEX field experiment.
Proc. Fourth Int. Winds Workshop, Saanenmoeser, Switzerland, EUMETSAT Publ. EUM-P-24, 279–280
.
Rohn
,
M.
,
G.
Kelly
, and
R. W.
Saunders
,
1998
:
Experiments with atmospheric motion vectors at ECMWF.
Proc. Fourth Int. Winds Workshop, Saanenmoeser, Switzerland, EUMETSAT Publ. EUM-P-24, 139–146
.
Schmetz
,
J.
,
K.
Holmlund
,
J.
Hoffman
,
B.
Strauss
,
B.
Mason
,
V.
Gaertner
,
A.
Koch
, and
L.
van de Berg
,
1993
:
Operational Cloud-Motion Winds from Meteosat Infrared Images.
J. Appl. Meteor
,
32
,
1206
1225
.
Tomassini
,
M.
,
G.
Kelly
, and
R. W.
Saunders
,
1999
:
Use and impact of satellite atmospheric motion winds on ECMWF analyses and forecasts.
Mon. Wea. Rev
,
127
,
971
986
.

Footnotes

* Current affiliation: *Ifb AG, Cologne, Germany.

+ Current affiliation: Met Office, Bracknell, Berkshire, United Kingdom.

Corresponding author address: Michael Rohn, ifb AG, Neumarkt-Galerie Neumarkt 2, 50667 Köln, Germany. Email: michael.rohn@ifbAG.com